Skip to content
AspectPNR confirmation signal
Methodology

How the prediction is made

No black box. Aspect follows a transparent four-step method that always leans on real outcomes first and tells you how much history sits behind every number.

What goes in

Three kinds of input

A prediction combines your specific booking with what has actually happened to similar tickets before. Nothing here is invented — where a detail is missing, we say so.

Your booking

Train, class, quota, and your status — confirmed, RAC or waitlisted — plus the waitlist type and your position in the queue.

Historical outcomes

How tickets on the same train, class, quota and waitlist type have actually resolved by departure, bucketed by position and days-to-go.

Timing

How far you are from departure, since confirmations cluster as the reservation chart nears and cancellation charges step up.

The four-step method

We use the best evidence available, and grade it

The engine walks these steps in order and stops at the first one with enough evidence. Each step carries its own confidence band, so a number is never presented as more certain than it is.

  1. Exact match

    High confidence

    The strongest signal: how this exact combination has cleared before, for positions within five of yours. Used when there's a solid run of past journeys behind it.

    Same train · class · quota · waitlist type · position ±5 · sufficient sample

  2. Route match

    Medium confidence

    When the exact combination is thin, we widen to the same route, class and waitlist type — still real outcomes, just a broader pool.

    Same route · class · waitlist type

  3. Train-level average

    Low confidence

    If route history is also limited, we fall back to how this train has performed overall. A rough read, and labelled as one.

    Any historical outcomes for this train

  4. Model estimate

    Estimated

    With no usable history yet, a fine-tuned model provides an estimate — and if that's unavailable, a deterministic fallback does. Always clearly marked as an estimate, and replaced automatically once real outcomes exist.

    No historical samples · estimate cached and superseded by real data

Whatever the step, you always get a number — we never return a “no data available” dead end. The confidence band is how we’re honest about the difference.
From number to signal

The probability, then the read

The percentage maps to a railway signal aspect so you can act on it at a glance. A confirmed ticket short-circuits to a full green — there’s nothing left to predict.

  • green · 75–100%

    Clear road ahead

  • yellow · 40–74%

    Proceed with caution

  • red · Below 40%

    Odds are against you

The rules we hold to

What keeps a prediction honest

Real data beats estimates, always

The chain tries real historical steps first. The moment genuine outcomes exist for your combination, they replace any earlier estimate — automatically.

Missing details never count against you

If your class, quota or waitlist type is unknown, we treat it as unknown — never as a negative signal — and tell you what we assumed.

An estimate is labelled an estimate

Step-four results carry an explicit note and an “Estimated” band. We never dress a model guess up as settled history.

Limits & uncertainty

What a prediction can’t do

Confirmation depends on other passengers cancelling, right up to the moment the chart is prepared. We estimate that from patterns; we can’t see individual decisions, sudden quota changes, or operational disruptions. A high probability means the odds are in your favour — not that the outcome is settled.

Treat every result as a prediction, not a guarantee. Keep a backup until your status is final, especially on a waitlisted ticket.

Go deeper

See how we measure ourselves

Transparency doesn't stop at the method. Here's how we think about accuracy — honestly, including what we can't yet claim.